Re: pg_dump, pg_dumpall and data durability

On 03/04/2017 01:08 AM, Robert Haas wrote:
> On Thu, Mar 2, 2017 at 5:02 AM, Michael Paquier
> <michael(dot)paquier(at)gmail(dot)com> wrote:
>> On Thu, Mar 2, 2017 at 2:26 AM, David Steele <david(at)pgmasters(dot)net> wrote:
>>> This patch is in need of a committer. Any takers?
>>> I didn't see a lot of enthusiasm from committers on the thread
>> Stephen at least has showed interest.
>>
>>> so if nobody picks it up by the end of the CF I'm going to mark the patch RWF.
>> Yes, that makes sense. If no committer is willing to have a look at
>> code-level and/or has room for this patch then it is as good as
>> doomed. Except for bug fixes, I have a couple of other patches that
>> are piling up so they would likely get the same treatment. There is
>> nothing really we can do about that.
> Before we reach the question of whether committers have time to look
> at this, we should first consider the question of whether it has
> achieved consensus. I'll try to summarize the votes:
>
> Tom Lane: premise pretty dubious
> Robert Haas: do we even want this?
> Peter Eisentraut: I had voiced a similar concern [to Robert's] previously
> Albe Laurenz: I think we should have that
> Andres Freund: [Tom's counterproposal won't work]
> Robert Haas: [Andres has a good point, still nervous] I'm not sure
> there's any better alternative to what's being proposed, though.
> Fujii Masao: One idea is to provide the utility or extension which
> fsync's the specified files and directories
>
> Here's an attempt to translate those words into numerical votes. As
> per my usual practice, I will count the patch author as +1:
>
> Michael Paquier: +1
> Tom Lane: -1
> Peter Eisentraut: -1
> Albe Laurenz: +1
> Andres Freund: +1
> Robert Haas: +0.5
> Fujii Masao: -0.5
>
> So, my interpretation is that, out of 7 votes, we have -2.5 and +3.5.
> Perhaps that is a consensus for proceeding, but if so it's a pretty
> marginal one. I think the next step for this patch is to consider why
> we shouldn't, in lieu of what's proposed here, add a pg_fsync utility
> that fsyncs a file or recursively fsyncs a directory, ship that, and
> let people use it on their pg_dump files and/or base backups if they
> wish. I am not altogether convinced that's a better option, but I am
> also not altogether convinced that it's worse. Also, if anyone else
> wishes to vote, or if anyone to whom I've attributed a vote wishes to
> assign a numerical value to their vote other than the one I've
> assigned, please feel free.
>
> The issue with this patch isn't that nobody is interested, but that
> not everybody thinks it's a good idea.
>

This is really a pretty small patch all things considered, and pretty
low-risk (although I haven;t been threough the code in fine detail yet).
In the end I'm persuaded by Andres' point that there's actually no
practical alternative way to make sure the data is actually synced to disk.

If nobody else wants to pick it up I will, unless there is a strong
objection.

cheers

andrew

--
Andrew Dunstan https://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services